AI Promised a Shorter Work Week. Tech Employees Are Logging 90-Hour Sprints Instead

The Promise: AI Was Supposed to Give Everyone Their Time Back

For the last few years, the loudest voices in AI have made the same pitch: automate enough of the grind and the work week shrinks. A Google engineering director predicted a four-day work week by 2025. In 2026, OpenAI went further, formally urging other employers to pilot a 32-hour week at full pay, arguing that AI gains should be converted into time off rather than absorbed by employers. Anthropic has pointed to Claude's ability to work autonomously for hours at a stretch as proof that AI can carry a full workday's worth of tasks. Meta's Mark Zuckerberg has said AI tools now let far fewer people accomplish more than ever before.

According to a BBC investigation published in August 2026, the employees actually building and using these tools tell a very different story.

The Reality Inside AI Companies

A former OpenAI technical employee told the BBC that the four-day week the company was publicly recommending to others was never trialled internally during their time there. They described a culture of recurring crisis meetings, weekend catch-up work, and blunt performance reviews, and said they typically logged around 70 hours a week, well above what they'd worked at previous tech jobs. Multiple current employees at OpenAI and Anthropic told the BBC that sprint periods, stretches of intense work ahead of a launch, can run for weeks at a time and top 90 hours in a single seven-day period. Neither company responded to the BBC's request for comment.

Getting 'Drafted' Whether You Like It or Not

At Meta, workers described being pulled onto urgent AI initiatives without being asked, a practice colleagues nicknamed being "drafted." One former employee put it bluntly: "You can't say no – or if you do, you have to quit." Those reassigned said the hours ran late into the night and through weekends, with an expectation of being reachable even when officially off the clock. Meta has since eased its hardline stance on mandatory reassignment, but only after many employees had already been moved. The company has also cut roughly one in ten roles this year, even as CEO Dario Amodei of Anthropic has separately warned that rising AI productivity could bring broader job losses across the industry.

It's Not Just the People Building AI

The strain isn't limited to AI labs. A former Google engineer, Amin Shali, told the BBC he left the company in May after his workload became dominated by overnight firefighting, a problem he linked to internal compute and infrastructure being redirected toward AI projects. He said his sleep and health improved noticeably once he left, and that heavy internal reliance on AI tooling had created a culture of constant pressure rather than relief.

Why the Time Savings Never Seem to Show Up

Research is starting to back up what these accounts suggest. An eight-month study of workers at a US tech company, run by researchers at UC Berkeley, found that employees given broad access to generative AI worked faster, took on a wider range of tasks, and ended up spending more hours working, not fewer, in part because of the extra time spent checking AI-generated output. MIT innovation researcher Neil Thompson has offered a similar explanation: even genuine efficiency gains tend to get absorbed by the work of rolling tools out and making sure they function correctly, and any hours that are freed up are often refilled with new work rather than banked as time off.

It's worth noting this isn't inevitable. Where companies have actually committed to a real 32-hour week, rather than just talking about one, the results look very different: research summarized by Forbes points to lower burnout, better work-life balance, and even productivity gains. The difference seems to come down to whether leadership treats the shorter week as a structural commitment or a talking point.

What This Means If You're Job Hunting in Tech or AI

If a role touches AI development in any way, it's worth treating "we're all about work-life balance" as a claim to verify, not a fact to accept. A few things worth doing before you sign anything:

• Ask direct questions in the interview about typical weekly hours during and outside of launch periods, not just the stated policy
• Ask specifically whether employees can be reassigned to urgent projects without being able to opt out
• Look for how recently any flexible-hours or four-day policy was actually used, not just announced
• Talk to current or recent employees where possible, since public statements and internal practice can diverge significantly

You can browse open roles with transparent hour expectations on HireDoor, and check verified company culture reviews before you accept an offer. For a deeper walkthrough of the questions worth asking, see our guide on evaluating work-life balance before you accept a job offer.

The Bottom Line

There's no US federal cap on weekly working hours for adults, unlike the UK and EU, where a 48-hour limit including overtime applies. That makes the gap between what AI companies say publicly and what their own employees report internally more than a messaging problem, it's something job seekers need to actively check for themselves rather than take on faith.